Seismic Data Denoising and Interpolation Using Deep Learning

Summary Seismic data recovery, including noise removal and interpolation, is virtual to improve data quality. We present a modified CBD-RDN network to remove noise and improve resolution simultaneously. As the performance of neural network is heavily influenced by the quality and diversity of data, we introduce two strategies, consistence of frequency bands and data augmentation. Numerical experiments on synthetic and field seismic data indicate that our method preserves more subtle features compared with traditional methods.